1. Neuroscience
Download icon

A neural circuit for gamma-band coherence across the retinotopic map in mouse visual cortex

  1. Richard Hakim
  2. Kiarash Shamardani
  3. Hillel Adesnik  Is a corresponding author
  1. University of California, United States
Research Article
  • Cited 3
  • Views 1,537
  • Annotations
Cite this article as: eLife 2018;7:e28569 doi: 10.7554/eLife.28569

Abstract

Cortical gamma oscillations have been implicated in a variety of cognitive, behavioral, and circuit-level phenomena. However, the circuit mechanisms of gamma-band generation and synchronization across cortical space remain uncertain. Using optogenetic patterned illumination in acute brain slices of mouse visual cortex, we define a circuit composed of layer 2/3 (L2/3) pyramidal cells and somatostatin (SOM) interneurons that phase-locks ensembles across the retinotopic map. The network oscillations generated here emerge from non-periodic stimuli, and are stimulus size-dependent, coherent across cortical space, narrow band (30 Hz), and depend on SOM neuron but not parvalbumin (PV) neuron activity; similar to visually induced gamma oscillations observed in vivo. Gamma oscillations generated in separate cortical locations exhibited high coherence as far apart as 850 μm, and lateral gamma entrainment depended on SOM neuron activity. These data identify a circuit that is sufficient to mediate long-range gamma-band coherence in the primary visual cortex.

https://doi.org/10.7554/eLife.28569.001

Introduction

Neural oscillations are ubiquitous features of brain activity and have been correlated with a diverse range of sensory, cognitive, and motor functions (Buzsáki and Draguhn, 2004; Jia and Kohn, 2011). It has been proposed that gamma oscillations (20–80 Hz) within the primary visual cortex (V1) synchronize the activity of functionally related neural ensembles, resulting in their selective amplification in downstream visual areas (Fries, 2009; Fries et al., 2007; Singer and Gray, 1995). In cats, gamma oscillations recorded in two spatially separated regions in V1 become phase-locked when a continuous visual object occupies their corresponding receptive fields (Gray et al., 1989; Gray and Singer, 1989). In primates, studies have observed strong correlations between gamma power and visual attention (Fries et al., 2001), as well as increased gamma-band synchrony among visual cortical areas during visual tasks, suggesting a role for gamma oscillations in visual perception and selective attention (Siegel et al., 2008). However, the precise function of gamma rhythms in visual processing remains controversial, with some studies providing evidence for a role in perceptual coding (Fries, 2009; Pritchett et al., 2015; Salinas and Sejnowski, 2001; Singer and Gray, 1995; Varela et al., 2001; Womelsdorf et al., 2007) and others arguing that neural codes based on the synchrony or phase of spike times during gamma oscillations are unlikely to contribute to visual processing (Ray and Maunsell, 2010; Shadlen and Movshon, 1999; Thiele and Stoner, 2003). Since the cellular and circuit mechanisms underlying long-distance gamma-band coherence are unknown, the causal manipulations that could assess whether gamma rhythms are required for key aspects of perception and cognition have yet to be defined.

Mechanistically, many different kinds of neural circuits may contribute to gamma-band synchronization (Bartos et al., 2007; Buzsáki and Wang, 2012; Mann and Paulsen, 2005; Whittington and Traub, 2003). In the mammalian forebrain, gamma oscillation generation is typically thought to depend on the reciprocal interaction between nearby recurrently connected excitatory and inhibitory neurons, with the precise timing of oscillations depending on the rapid recruitment of inhibitory neurons to transiently and periodically suppress local network activity (Buzsáki and Wang, 2012; Cardin, 2016; Jefferys et al., 1996; Mann et al., 2005; Tiesinga and Sejnowski, 2009). Several studies have correlated the activity of individual subtypes of inhibitory cortical neurons with gamma-paced firing, finding, for instance, that fast-spiking parvalbumin-positive (PV) basket cells phase-lock particularly strongly to gamma-band activity in the local field potential (Bartos et al., 2007; Buzsáki, 2004; Csicsvari et al., 2003; Gulyás et al., 2010; Hasenstaub et al., 2005; Salkoff et al., 2015; Siegle et al., 2014). Other studies have demonstrated that different or multiple inhibitory subtypes are involved in gamma generation and coherence (Craig and McBain, 2015; Kipiani, 2009; Takada et al., 2014; Vierling-Claassen et al., 2010; Whittington et al., 2011), and that gap junctions might also contribute (Ainsworth et al., 2011; Long et al., 2005; Szabadics et al., 2001; Traub et al., 2001). Recently, it was also demonstrated that, in mouse primary visual cortex, somatostatin-positive (SOM) interneurons contribute to stimulus-size dependent, visually induced oscillations and coherence (Chen et al., 2017; Veit et al., 2017). Moreover, theoretical work has also suggested that dendrite-targeting interneurons might contribute to gamma-band coherence in the hippocampus (Tort et al., 2007).

Despite this work, the cellular and synaptic basis for gamma coherence across the retinotopic map in the visual cortex is poorly understood. This may be due to technical challenges associated with manipulating and recording the activity of neuronal ensembles separated by large distances with the sufficient spatial and temporal control needed to probe the underlying mechanisms of neural synchronization. Previously, we and others have shown that optogenetic stimulation of L2/3 pyramidal cells (PCs) induces powerful gamma-band oscillations in cortical slices and in vivo (Adesnik and Scanziani, 2010; Shao et al., 2013; Takada et al., 2014). In this study, we employed patterned illumination and bidirectional optogenetics to precisely control the generation and entrainment of gamma rhythms across the retinotopic map of the mouse primary visual cortex.

Results

Long-range coherence of photo-induced gamma rhythms

We took advantage of the ability to reliably and potently generate gamma oscillations by optogenetically stimulating PCs in L2/3 of cortical slices (Adesnik and Scanziani, 2010, 2012; Shao et al., 2013; Takada et al., 2014). ChR2 was transfected into a sparse subset of L2/3 PCs in V1 using in utero electroporation (Saito and Nakatsuji, 2001), and a slow ramp of blue light targeted to L2/3 was used to reliably drive oscillatory network activity. Under these conditions, ChR2 expression is restricted to excitatory neurons (Figure 1—figure supplement 1) (Adesnik and Scanziani, 2010), and therefore all optogenetically evoked inhibition is driven polysynaptically through the network, rather than being of monosynaptic origin. Consistent with prior work in both S1 and V1, wide-field illumination of L2/3 generates strong gamma rhythms in excitatory and inhibitory currents measured in L2/3 cortical neurons (Figure 1A,B). To gain control over the spatial profile of excitation, we built and characterized a digital-micromirror-device (DMD) based illumination system that generates arbitrary multicolor light patterns with high spatial and temporal precision (Figure 1—figure supplement 2, Figure 4—figure supplement 1). Using this system, we found that the power of the gamma oscillations depended on the area of illumination, reminiscent of the dependence of gamma oscillations on visual stimulus size in vivo (Gieselmann and Thiele, 2008; Jia et al., 2013; Ray et al., 2013; Veit et al., 2017) (Figure 1C. Analyzed from 0 to 1000 ms post-stimulus onset.).

Figure 1 with 2 supplements see all
Horizontal circuits recruit local SOM interneurons to synchronize distant gamma generators.

(A) Experimental schematic: A ChR2-negative Pyramidal cell is recorded in L2/3 of V1 while other ChR2-expressing L2/3 neurons are photo-stimulated with different sizes of blue light stimuli using a digital-micromirror-device (DMD). (B) Top: Time course of the light stimulus intensity (final intensity 1.1 mW/mm2, see Materials and methods). Bottom: Example traces of voltage-clamped excitatory postsynaptic current (EPSC, red) and inhibitory postsynaptic current (IPSC, blue) during photo-induced gamma rhythms in V1. (C) Plot of peak gamma power versus the width of the photo-stimulus on L2/3 (n = 8, p<10−4, Kruskal-Wallis ANOVA). Errorbars are s.e.m. (D) Experimental schematic: two ChR2-negative L2/3 pyramidal cells are simultaneously recorded while nearby ChR2-expressing L2/3 PCs are focally activated with separate blue light patches using a digital micro-mirror device (DMD). The distance between the blue light patches ranged from 275 to 850 μm (see Figure 1—figure supplement 1B). (E) Example traces of the voltage-clamped IPSCs from a pair of simultaneously recorded L2/3 PCs during photo-induction of two separate gamma oscillations. (F) Oscillation-triggered average of the IPSCs recorded in the pair in B) (triggered off the oscillations in one of the two cells, labeled in dark blue). Shading represents one standard deviation. (G–I) As in (D–F) but following a transection of L2/3 between the two recorded L2/3 PCs in transfected slices. (J) Scatter plot of the peak coherence of the oscillations in the two recorded neurons between the cut and the two intact conditions. Mean peak coherence with 275–400 μm separation (close): 0.72 ± 0.04, n = 6 pairs; mean peak coherence at 625–850 μm separation (far): 0.44 ± 0.09, n = 7 pairs; mean peak coherence at 275–400 μm with L2/3 cut (cut): 0.11 ± 0.01, n = 11 pairs; p<10−3, Wilcoxon rank sum test between close and cut conditions; p<10−3, Wilcoxon rank sum test between far and cut conditions. Errorbars are s.e.m.

https://doi.org/10.7554/eLife.28569.002

To probe the mechanisms of long-range gamma coherence across the retinotopic map, we examined if photo-induced gamma rhythms in L2/3 are phase-locked across distant ensembles. Using the DMD, we generated two patches of blue light, separated by distances ranging from 275 to 850 μm, to activate two separate regions in L2/3 of V1. This was done while making simultaneous intracellular recordings from two ChR2-negative PCs located at the center of each patch of light (Figure 1D–F). Photo-stimulation generated gamma oscillations in each illuminated region, as measured by large oscillatory currents recorded in each cell, and when both sites were co-stimulated, the two oscillations strongly phase-locked (Figure 1D–F). Coherence persisted even when the recorded cells were separated by up to 850 μm (Figure 1J), demonstrating that photo-induced gamma rhythms can synchronize over long distances.

L2/3 PCs project long-range horizontal axons that traverse the retinotopic map in V1 (Gilbert and Wiesel, 1989) and represent a plausible substrate for synchronizing gamma activity across spatially distant neural ensembles (Gray et al., 1989). To determine if horizontal axons in L2/3 are required for the long-range gamma phase-locking, we made a vertical cut between the two recorded cells, severing L2/3 horizontal axons. This transection spared gamma rhythm generation on either side of the cut, but abolished phase-locking between the two areas (Figure 1G–J; n = 11 pairs, p<10−3, Wilcoxon rank sum test, see figure legend for statistics). These data demonstrate that lateral connections across L2/3 in V1 are necessary for long-range intralaminar gamma coherence.

L2/3-induced gamma rhythm generation depends on somatostatin interneurons

To address the possible mechanisms of gamma rhythm induction, we probed the activity of the two major subclasses of cortical interneurons that provide inhibition to L2/3 PCs. Although substantial evidence implicates PV neurons in gamma entrainment (Buzsáki and Wang, 2012; Cardin et al., 2009; Sohal et al., 2009), we previously demonstrated that horizontal axons in L2/3 preferentially recruit SOM neurons in vitro (Adesnik et al., 2012), and that SOM neurons in vivo are also involved in visually induced gamma rhythms (Veit et al., 2017). Therefore, we targeted PV and SOM neurons to measure their activity during photo-induced gamma oscillations. To this end, we expressed ChR2 in PCs via in utero electroporation in SOM-Cre and PV-Cre mice crossed to a tdTomato Cre-reporter strain (Figure 1—figure supplement 1). This permitted photo-stimulation of L2/3 PCs, as in previous experiments, but further allowed simultaneous targeted electrophysiological recordings from tdTomato-expressing PV or SOM neurons. Strikingly, targeted loose-patch recordings from SOM and PV neurons in brain slices demonstrated that SOM neurons were reliably driven and tightly correlated with the photo-induced gamma rhythm in L2/3 (Figure 2A–C, n = 10 cells, mean induced rate: 25 ± 5 Hz. Data analyzed from 0 to 1000 ms post-stimulation onset.). Under the same conditions, the majority of L2/3 PV neurons fired at comparatively low rates or did not fire at all (4 ± 2 Hz, p=0.003, Wilicoxon rank sum test). To probe the basis for this unexpected result, whole-cell recordings from SOM neurons were made and demonstrated that they receive powerful, rhythmic excitation but very little inhibition (Figure 2D,E). In contrast, PV neurons and PCs received much more synaptic inhibition, consistent with their known inputs from SOM cells (Figure 2D,E; see figure legend for statistics) (Hioki et al., 2013; Pfeffer et al., 2013). The large difference in their excitation/inhibition ratio provides a synaptic basis for the differential recruitment of SOM and PV neurons during photo-induced gamma.

L2/3 gamma rhythms recruit SOM inhibitory neurons.

(A) Top: Schematic of experiment. Loose-patch recordings are made from PV or SOM cells, and whole-cell recordings are made from Pyramidal cells (PC) to correlate to gamma rhythms. Slices expressing ChR2 in L2/3 and tdTomato in PV or SOM cells (see Figure 1—figure supplement 1). Bottom: example loose-patch recordings from a SOM and a PV neuron during photo-induced gamma. (B) Phase histogram of the spikes recorded in a SOM neuron during photo-induced gamma overlaid with the simultaneously recorded cycle-averaged IPSC. (C) Plot of the mean firing rate of PV and SOM neurons during photo-induced gamma (PV: n = 11 cells, SOM n = 10 cells, p<0.01, Wilcoxon rank sum test). Baseline spike rate was zero. (D) Representative traces of synaptic excitation (red) and inhibition (blue) from a SOM, PV, and PC. Baseline synaptic input is near-zero for both EPSCs and IPSCs. (E) Plot of the average E/I ratio in SOMs, PVs, and PCs during gamma activity in brain slices (n = 12 SOM cells, mean E/I ratio: 2.5 ± 0.5; 8 PV cells, mean E/I ratio: 0.43 ± 0.09; 10 pyramidal cells, mean E/I ratio: 0.054 ± 0.010; p<10−3, Wilcoxon rank sum test between SOM and PV; p<10−5, Kruskal-Wallis ANOVA). Error bars are s.e.m.

https://doi.org/10.7554/eLife.28569.005

To causally dissect the specific contributions of SOM neurons to photo-induced gamma rhythms, we optogenetically suppressed SOM neurons and measured the resulting impact on gamma-band synaptic activity. To achieve simultaneous activation of PCs and suppression of SOM neurons, we injected a Cre-dependent adeno-associated virus (AAV) driving a red-light activated neural silencer, eNpHR3.0, into the visual cortex of in utero electroporated SOM-Cre mice. During photo-stimulation of PCs to induce gamma rhythms with blue light, red light was added to silence SOM neurons (blue light from 0 to 1000 ms post-stimulation onset, red light from 250 to 750 ms; analysis conducted from 250 to 750 ms). Local suppression of SOM neurons strongly reduced the inhibitory gamma power observed in PCs during photo-induced gamma activity (Figure 3A,B mean power reduction: 74 ± 3.5%, p<10−5, Wilcoxon signed-rank test). Suppressing SOM cells also desynchronized excitatory synaptic input, demonstrating that their output is critical for pacing the activity of the excitatory network in the gamma band (Figure 3B,C; n = 8 cells, mean power reduction: 63 ± 7%, p=0.004, Wilcoxon signed-rank test). In contrast, suppressing PV neurons had no significant effect on gamma power (Figure 3—figure supplement 1A–E, n = 9 cells, mean power reduction: 2.5 ± 6%, p=0.25, Wilcoxon signed-rank test), while it did significantly reduce the disynaptic IPSC to a brief pulse of blue light (Figure 3—figure supplement 1F, p=0.016, Wilcoxon signed-rank test). This demonstrates that gamma generation within L2/3 critically depends on the activity of SOM neurons.

Figure 3 with 2 supplements see all
L2/3 gamma rhythms depend on SOM neuron activity.

(A) Schematic of experiment. Whole-cell recordings are made from L2/3 Pyramidal cells (PC) in slices containing ChR2 in L2/3 and eNpHR3.0 in SOM cells. In some experiments a SOM cell was simultaneously recorded. (B) Left: loose-patch recording of spiking in an example eNphR3.0 expressing-SOM cell (black, top trace) and the photo-induced excitatory (red) and inhibitory (blue) currents in a representative PC. Right: Same cells, but with the addition of red light (red bar) to suppress SOM cells. Bottom: expanded excitatory currents. (C) Power spectra of the recorded inhibition (top) and excitation (bottom) in the PC in B) under control conditions (black) and during photo-suppression of SOM cells (red). (D) Scatter plots of the peak gamma power of inhibition (top) and excitation (bottom) with and without photo-suppression of SOM cells (Inhibition: n = 24, p<10−5, Wilcoxon signed rank test) (Excitation: n = 9, p<0.01, Wilcoxon signed rank test).

https://doi.org/10.7554/eLife.28569.006

These experiments still leave open the possibility that SOM neurons are only critical for maintaining gamma entrainment, but are not required at the very initiation of the photo-induced gamma oscillations, which might instead depend on PV neurons. This possibility is made more likely by studies showing that excitatory synapses onto PV neurons are strong but highly adapting, while those onto SOM cells are initially weak but highly facilitating (Markram et al., 2004; Reyes et al., 1998). We, therefore, tested whether optogenetically suppressing SOM neurons at the induction of gamma oscillations would also abolish gamma rhythms. We found that suppressing SOM cells from before gamma induction also nearly completely abolished gamma rhythms, indicating that they are required even at the very earliest stages of gamma induction (Figure 3—figure supplement 2F–H). These results are consistent with direct recording of the spiking of SOM and PV cells during the time course of photo-induced gamma. We found that even at the early stages of gamma, SOM neuron activity still far exceeded that of PV neurons (Figure 3—figure supplement 2A–E). Although these data do not exclude the possibility that PV neurons could be necessary for oscillatory activity in response to other photo-stimulation protocols, they demonstrate that SOM neurons are necessary for both the initiation and maintenance of L2/3-induced gamma oscillations studied here.

Lateral entrainment of pyramidal cells requires SOM activity

Whereas the preceding experiments establish that SOM neurons that are local to the oscillating ensemble are required for maintaining photo-induced gamma oscillations, the circuits that couple distal ensembles remain uncertain. Therefore, we tested whether SOM neurons are also involved in phase-locking spatially separated ensembles. We hypothesized that the horizontally projecting axons of L2/3 PCs might synchronize gamma rhythms across the retinotopic axis of the cortex by synapsing onto laterally located SOM cells, which then enforce phase-locking through their inhibitory output. To test this hypothesis, we optogenetically induced gamma activity in one region of the slice with a blue light patch, while inducing action-potentials in a ChR2-negative PC in different region of the slice using whole-cell somatic current injection (Figure 4A,B and Figure 4—figure supplement 1). We used current injection in the distal PC to evoke action potentials since the horizontal synaptic input from photo-stimulated L2/3 PCs tends to exert a strong net suppressive effect and does not evoke spiking on its own. The distance between the distal current-injected pyramidal cell and the border of the blue light was kept to approximately 325 μm, which maintained robust spike-oscillation entrainment (Figure 4B–C). The blue light region extended 600 μm laterally (950 μm maximum separation between distal PC and blue light region, see Figure 4—figure supplement 1C). Therefore, under these conditions, if suppressing SOM cells local to the current-injected pyramidal cell were to abolish gamma entrainment, it would support the notion that local SOM cell activity is required for both local and long-range gamma band synchronization.

Figure 4 with 2 supplements see all
Gamma band synchronization across the retinotopic axis depends on local SOM neuron activity.

(A) Schematic of experiment: One area of the slice is photo-stimulated with blue light to generate a local gamma oscillation, while the second distal site is simultaneously illuminated with red light to suppress local SOM activity focally (see Figure 4—figure supplement 1C). (B) Example traces from a simultaneously recorded pair of L2/3 PCs. The cell in the blue light zone is recorded in voltage clamp (blue, IPSC), while the cell in the red light zone is recorded in current clamp and stimulated with a current injection to evoke action potentials (gray). Current injections ranged between 300 and 600 pA. Bottom: Expanded trace showing spikes occurring at the toughs of inhibition. (C) Population phase histogram of the spikes of the current injected L2/3 PC in the red light zone under control conditions (blue light only) and during photo-suppression of local SOM cells relative to the phase of the gamma oscillation recorded in the second PC in the blue light zone. Error bars are s.e.m. (D) Scatter plot of the pairwise phase consistency of the L2/3 PC recorded in the red light zone relative to the oscillation recorded in the L2/3 PC in the blue light zone (p<10−7, n = 8 pairs, Wilcoxon signed-rank test).

https://doi.org/10.7554/eLife.28569.009

In control conditions, the spiking of the current-injected neuron strongly phase-locked to the distally generated gamma oscillations (Figure 4C). In interleaved trials, we inactivated SOM cells specifically near the distal target PC with a patch of red light that did not overlap with the gamma oscillation initiation zone (see schematic, Figure 4A, and Figure 4—figure supplement 1). Optogenetic suppression of local SOM neurons with red light abolished phase-locking between PC-spikes and the distally generated gamma oscillation (Figure 4C,D, n = 8 pairs, PPC control: 0.120 ± 0.008; PPC red light: 0.004 ± 0.007, p=0.008, Wilcoxon signed-rank test. Data analyzed from 200 to 1000 ms post-stimulation onset.). Together, these data support the notion that the activity of SOM neurons outside of the gamma initiation zone is critical for phase-locking spatially separate oscillating ensembles across L2/3 of V1 in brain slices at least 325 μm apart. However, it should be noted that direct optogenetic suppression of SOM terminals, and partial suppression of SOM neurons between the blue and red light zones might also contribute (see Figure 4—figure supplement 1F)

In our first line of experiments (Figure 1), we observed significant coherence even for ensembles 625–850 μm apart, albeit less so compared to ensembles 275–400 μm apart. This long-range coherence, and its dependence on distance, could be explained by multiple mechanisms: long-range L2/3 excitatory axons to distal SOM neurons, mid-range L2/3 excitatory axons to SOM neurons located between the two ensembles and horizontal projections from these same SOM neurons, recurrently propagating loops of neural synchronization, or electrical coupling between SOM neurons that could facilitate synchronization across the retinotopic axis of V1. To begin to address these possibilities, we mapped the range of excitatory input to SOM neurons from PCs by patching SOM neurons and stimulating increasingly more distant ensembles of L2/3 PCs (Figure 4—figure supplement 2A). We found that SOM neurons received significant, albeit attenuated excitatory input from L2/3 PCs as far away as 450 μm (Figure 4—figure supplement 2B,C, average maximum excitatory current: 120 ± 90 pA, 6 ± 2% of the maximum, n = 9). In current clamp, the same neurons reached an average maximal depolarization at 450 μm (calculated in in the final 100 ms of the photo-stimulus) of 7 ± 5 mV (n = 7), which was rarely enough to drive spiking on its own. At the same time, inhibitory currents in L2/3 PCs decayed to 4 ± 1% of their maximum at the same 450 μm distance (Figure 4—figure supplement 2B,C, maximum inhibitory current 20 ± 10 pA, n = 8).

When recording SOM neurons, we did not observe any clear spikelets or evidence of subthreshold electrical coupling that might mediate signal propagation between SOM neurons across the retinotopic access of the cortex (Figure 4—figure supplement 2D–G). However, it is possible that spikelets would be temporally filtered and therefore indistinguishable from EPSPs or EPSCs (Hu and Agmon, 2015), particularly against the large barrages of glutamatergic inputs SOM cells receive; thus the contribution of gap junctional coupling requires further exploration. When we examined the spatial spread of disynaptic inhibition onto PCs, inhibition exhibited a similar spatial decay as that of excitatory input to SOM neurons (Figure 4—figure supplement 2B,C). These data are consistent with a mechanism for long-range coherence involving the summed horizontal dendritic and axonal fields of pyramidal and SOM neurons.

Discussion

Using patterned illumination optogenetics and multiple targeted intracellular recordings from various cortical neuron subtypes, this study establishes a novel circuit that is sufficient for synchronizing independently oscillating neuronal ensembles in the gamma band. Independent oscillating ensembles, in separate patches of V1, mutually connect to each other via disynaptic circuits composed of horizontally projecting excitatory axons of L2/3 neurons and SOM neurons.

These data support the conclusion that the gamma oscillations we observed here are generated locally by a PING (Pyramidal-Interneuron-Gamma) circuit motif (Buzsáki and Wang, 2012; Tiesinga and Sejnowski, 2009) between pyramidal neurons and cortical inhibitory neurons, with SOM neurons playing a critical role. Because the oscillation is induced by the selective photo-stimulation of excitatory neurons, and since SOM neurons largely lack recurrent synaptic connectivity in L2/3 (Pfeffer et al., 2013), an ING (Interneuron-Gamma) model appears unlikely. In vivo, visually induced gamma rhythms in mouse V1 are highly sensitive to SOM neuron suppression, whereas PV activity appears to be necessary for circuit stabilization (Veit et al., 2017). Taken together with previous work showing that L2/3 horizontal axons preferentially recruit SOM neurons (Adesnik et al., 2012), the data presented here support a model for phase-locking spatially separated ensembles across V1. Gamma rhythms, generated locally by a PING network in independent patches of V1, mutually connect to each other via circuits composed of horizontally projecting excitatory axons of L2/3 neurons and SOM neurons, which can extend the range of synchronization through their own horizontal projections. The recruited SOM inhibition synchronizes the two ensembles, even over relatively long distances. Although it remains to be further explored, since SOM interneurons are known to be gap junction-coupled (Gibson et al., 1999; Hu and Agmon, 2015), electrical connectivity among the SOM population might facilitate or reinforce gamma rhythm generation. We did not observe clear indications of spikelets in SOM neurons during horizontally propagating gamma oscillations that would support this hypothesis, although temporal filtering of spikelets could make distinguishing them from normal glutamatergic EPSPs challenging. Additionally, the impact of VIP neurons, which are particularly abundant in L2/3 and project strongly to SOM cells (Pfeffer et al., 2013), may also influence gamma entrainment by SOM cells, possibly in state-dependent manner (Fu et al., 2014).

Using circuit mapping approaches, we explored the mechanisms underlying our observation that oscillating ensembles exhibit significant coherence across long distances (>800 μm) across the retinotopic axis of V1. When we recorded from SOM cells and mapped their horizontal excitatory input from one side, by 450 μm excitatory input to SOM neurons was 6 ± 2%of its maximum value. Although in most cases this input was insufficient to drive any spikes on its own, it still led to significant depolarization of the membrane potential. These measurements probably underestimate the true range of input to SOM neurons due to the sectioning of axons during brain slice preparation, but they do not support the notion that horizontal axons from the directly photo-stimulated L2/3 PCs are sufficient to explain the long-distance entrainment we observe beyond 450 μm. Our results most closely align with the model that SOM cells located between distant ensembles are driven by the sum of the input from the two sources, and synchronize the two ensembles through their own medium range projections. Interestingly, this architecture potentially resembles a phase-locked loop control system, with SOM neurons acting as phase comparators (Ahissar, 1998). For more closely spaced ensembles (<450 μm apart), horizontal input from each PC ensemble is sufficient to directly drive spiking of SOM cells local to the other ensemble.

In experiments using spatially restricted red light to inhibit SOM neurons near a phase-locked PC (Figure 4), it is possible that some of the reduction in phase-locking could be due to direct suppression of long-range SOM cell axons originating from the gamma oscillation initiation zone, or to direct somato-dendritic suppression of SOM cells located between the blue and red light zones. Synaptic terminal suppression by eNpHR3.0, though possible here, appears to be effective only early in an action potential train, at least in thalamic axons in which it was tested (Mahn et al., 2016). Since, during gamma rhythms, SOM cells fire prolonged spike trains up to 30 Hz, it seems unlikely that the effects we observed would be due to synaptic terminal suppression per se since illuminated axons would be able to maintain transmitter release after the first action potential. Alternatively, somato-dendritic hyperpolarization of SOM neurons located between the blue light and red light zone is plausible. While only a minority of SOM cells in L2/3 of V1 possess long-range horizontal axonal projections (McGarry et al., 2010), as suggested above, these few SOM neurons might be sufficient to help mediate the gamma band coherence.

While our data demonstrate the critical role SOM cells play in horizontally propagating gamma rhythms, in many circuits, both in vivo and in vitro, PV neurons also fire and synchronize to gamma rhythms, and in some cases, are necessary or sufficient to drive gamma rhythmicity (Sohal et al., 2009; Cardin et al., 2009). Several factors might account for the apparent contradiction with the previous literature. First, unlike these two previous studies where PV neurons were manipulated directly, we chose to directly activate excitatory neurons and allow the downstream circuit architecture to dictate the resulting network activity and dynamics. This suggests that while PV neurons or other mechanisms may be sufficient to drive gamma rhythmicity, the network architecture of L2/3 preferentially recruits SOM cells to fulfill this role. Moreover, recent results have shown that repeating the exact experiments of these two optogenetic studies with SOM cells could reproduce the same effects, demonstrating that SOM neuron activity is likewise sufficient to entrain gamma on its own (Veit et al., 2017)(Chen et al., 2017).

Second, gamma rhythms are phenomenologically diverse, exhibiting a range of peak frequencies, receptive fields, neuron subtype spike-time coupling, and other features (Buzsáki and Wang, 2012; Colgin et al., 2009; Gieselmann and Thiele, 2008; Niell and Stryker, 2010; Ray and Maunsell, 2010; Saleem et al., 2017; Sohal et al., 2009). This diversity strongly suggests diverse underlying mechanisms of oscillation generation and modulation across brain areas and network states. For this reason, care should be taken to compartmentalize conclusions on the underlying mechanisms of gamma generation based on differences in phenomenology and experimental approach.

Third, it is possible that in previous studies that observed tight coupling of PV (or ‘fast-spiking’ putative PV neurons) to ongoing gamma rhythms, the coupling is actually a result of gamma entrainment by SOM cells (which synapse strongly onto PV neurons) and not due to the PV neurons generating the gamma rhythm themselves. In mouse V1, PV neurons strongly couple to visually induced gamma, but optogenetically suppressing SOM, but not PV neurons, disrupts gamma rhythms in vivo (Veit et al., 2017), further suggesting this possibility.

Fourth, although our data demonstrate that SOM neurons are required for both local gamma generation and driving gamma-band coherence across the horizontal axis of the cortex, our data do not exclude a potential role for PV and other interneuron subtypes, such as VIP cells or deep-layer Martinotti cells. During gamma generation, PV neuron spikes, though sparse, are phase-locked to the gamma rhythm, and as a population may help reinforce the entrainment of local and/or distal excitatory neurons. This possibility is balanced with previous data showing that increasing stimulus sizes trend towards SOM neuron recruitment and away from PV neurons recruitment in vivo (Adesnik et al., 2012), which is in agreement with our findings in vitro, and our data showing that inhibition of distal SOM cells essentially abolishes distal entrainment to gamma rhythms. Moreover, although the oscillations we observed were strongly dependent on horizontal connections across L2/3 only (Figure 1G), gamma rhythm generation could involve intracolumnar SOM cells in lower layers as well (Kapfer et al., 2007).

Horizontal circuits that engage SOM neurons are known to be important for contextual modulation in V1 (Adesnik et al., 2012), and recent data also show them to be critical for size-dependent visually induced gamma rhythms in mouse V1 (Veit et al., 2017), and in the olfactory bulb (Lepousez et al., 2010). Moreover, since SOM cells target PC dendrites, their inhibitory action during gamma oscillations could have significant impact on spike timing dependent plasticity in synapses located on distal dendrites. Taken together with our findings that SOM neurons are sufficient for gamma generation within L2/3 and drive gamma-band synchrony in distal neurons, SOM neurons are emerging as critical contributors to both gamma generation and long-range coherence in the cortex. Moreover, these findings suggest a new route to probe functional roles of gamma oscillations and gamma-band coherence in visual perception: direct manipulation of SOM neurons or their synaptic partners.

Materials and methods

Transgenic mice

Request a detailed protocol

All experiments were performed in accordance with the guidelines and regulations of the ACUC of the University of California, Berkeley and the IACUC of the University of California, San Diego. Both female and male mice were used. In vitro experiments were performed on animals aged 21–30 days old. Mice strains used in this study were wild-type ICR mice (Charles River), SOM-IRES-Cre (JAX stock 013044), PV-Cre (JAX stock 008069), and Rosa-LSL-tdTomato (JAX stock 007909). Mice were subjected to in utero electroporation at E15-E16 as previously described (Adesnik et al., 2012), which exclusively labels L2/3 pyramidal cells.

Viral injection

Request a detailed protocol

Neonatal SOM-Cre mice (P1-3) were briefly cryo-anesthetized and placed in a head mold. Transcranial injection of ~45 nl of undiluted AAV9-DIO-Ef1a-eNpHR3.0-YFP (UPenn Vector Core) was performed using a Drummond Nanoject injector at three locations in V1. With respect to the lambda suture coordinates for V1 were 0.0 mm AP, 2.2 mm L and injection was 150–400 μm under the skull. For the experiments in Figure 3—figure supplement 1, the same virus was injected into juvenile PV-Cre mice (P14-P18) at 0.0 AP, 2.75 L, and 150–300 μm under the skull.

In utero electroporation

Request a detailed protocol

Timed pregnant wild-type ICR mice were either purchased directly or timed-matings were set up between female ICR white mice (Charles River) and homozygous male SOM-IRES-CRE;Rosa-LSL-tdTomato or PV-CRE;Rosa-LSL-tdTomato mice. Pregnant mice at E15-16 were anaesthetized with 2.0% isoflurane, the abdomen was cleaned with 70% ethanol and swabbed with iodine, and a small vertical incision was made in the skin and abdominal wall and 8–12 embryos gently exposed. Each embryo was injected with 0.5–1 μl of DNA solution and 0.05% Fast Green dye. pCAG-ChR2-Venus plasmid DNA was mixed with pCAG-GFP for a total of 1–2 μg ChR2 DNA and 0.5–1 μg of fluorophore DNA per injection. Alternatively, pCAG-ChR2-mCherry without pCAG-GFP was used. We used a pressure-controlled beveled glass pipette (Drummond, Custom Microbeveller) for injection. After each injection, the embryos were moistened with saline and voltage steps via tweezertrodes (BTX, 5 mm round, platinum, BTX electroporator) were applied with the positive electrode placed over the visual cortex and the negative electrode placed under the head of the embryo. Voltage was 40 V for 5 pulses at 1 Hz, each pulse lasting 50 ms. The embryos were returned to the abdomen, which was sutured, followed by suturing of the skin. The procedure typically lasted under 30 min. On the day of birth, animals were screened for location and strength of transfection by trans-cranial epifluorescence under an Olympus MVX10 fluorescence stereoscope.

In vitro recording

Request a detailed protocol

Mice were deeply anesthetized with isoflurane and quickly decapitated. 400 μm thick slices were cut on a microslicer (DTK-1000) and incubated at 34 degrees for 30–45 min, and then at room temperature in sucrose cutting solution (in mM: NaCl, 83; KCl, 2.5; MgSO4, 3.3; NaH2PO4, 1; NaHCO3, 26.2; D-glucose, 22; sucrose, 72; and CaCl2, 0.5, bubbled with 95% O2% and 5% CO2). A slice was transferred to a submerged chamber perfused with warmed (32 degree Celsius) ACSF (in mM: NaCl, 119; KCl, 2.5; NaH2PO4, 1.3; NaHCO3, 26; D-glucose, 20; MgCl2, 1.3; CaCl2, 2.5; and mOsm, 305, bubbled with 95% O2% and 5% CO2) and held down with nylon threads on a platinum harp. Excitatory and inhibitory currents were recorded in the voltage clamp mode with a cesium based internal solution (in mM: CsMeSO4, 115; NaCl, 4; HEPES, 10; Na3GTP, 0.3; MgATP, 4; EGTA, 0.3; QX-314-Cl, 2.5; BAPTA-5Cs, 10), and action potentials were recorded in a solution in which cesium was exchanged with potassium and QX-314 and BAPTA were omitted. Patch pipettes had resistances of 2–3 MOhm. Signals were amplified with two Multiclamp 700B amplifiers (Molecular Devices), filtered at 2 kHz and digitized via a National Instruments A/D card at 20 kHz. Custom software in Matlab (Mathworks) controlled all aspects of the experiment. Whole cell recordings were made from a L2/3 pyramidal shaped neurons. In some experiments, simultaneous recordings were made between two pyramidal cells or a pyramidal cell and a fluorescently labeled interneuron. Cells that exhibited direct photocurrents were discarded, except in Figure 1—figure supplement 2, where the spatial profile of photocurrents were measured. For whole-cell recordings, series resistance was less than 25 MOhm (uncompensated), and if this value changed in any cell by more than 20% during the course of an experiment, the cell was discarded. The light stimulus was always centered on layer 2/3 unless otherwise noted.

In Figure 4, current injection was used to bring the distal Pyramidal cell in current clamp to spike at a rate of ~10 Hz. Current injection amplitudes ranged from 300 to 600 pA, which corresponds to roughly 150–450 pA above typical L2/3 Pyramidal cell rheobase (~150 pA) (Guan et al., 2007; Lefort et al., 2009; van der Velden et al., 2012).

Optogenetic stimulation in vitro with a DMD

Request a detailed protocol

For testing size-dependence and spatial coherence of gamma-oscillations (Figure 1), blue light was generated using a 1W 445 nm diode laser (Ultralasers) and routed via a liquid light guide into a CEL5500 digital micromirror device (DMD) (Digital Light Innovations, Austin, Texas). The projection from the DMD was then collimated and integrated into the light path of the microscope, before being focused onto the slice chamber using a 5 × objective lens (Olympus). The width of the blue light regions was 300 μm in the horizontal axis, and the borders of the two regions were never separated by less than 150 μm when varying the distance between the two recorded neurons. Cells were patched roughly in the middle of their respective regions. Activity was strongly biased, but not completely restricted to the illuminated regions, potentially due to neuronal processes extending into the illuminated regions as well as light scatter in the tissue (Figure 1—figure supplement 2, Figure 4—figure supplement 1).

For Figure 2 and 3, blue and red light was generated using a multicolor LED light engine (Lumencor Spectra X, Beaverton, Oregon) controlled by digital outputs (NI PCIe-6353), and was then routed via a liquid light guide before being focused onto the slice chamber using a 5 × objective lens (Olympus, Japan). To generate ramps of blue light, the signal was frequency modulated (0–5 kHz). Optogenetic photocurrents and spiking followed the integral of the light power with no correlation to the frequency of the rapidly pulsed light.

For Figure 4, blue light was generated using a 5W 445 nm laser diode (Nichia NUBM44) and red light was generated using a 700 mW 638 nm laser diode (Oclaro HL63193MG). The red and blue beams were expanded and combined before passing directly into the CEL5500 digital micromirror device (DMD), then focused onto the slice chamber using a 5 × objective lens (Olympus). The width of the blue light region was 600 μm in the horizontal axis, the width of the red light region was 350 μm, the separation between the borders of the red and blue regions was 150 μm (Figure 4—figure supplement 1C), and cells were patched close to the middle of their respective regions. In order to achieve two color spatial light modulation simultaneously with a single DMD, we temporally multiplexed the two color channels. The spatial patterns for the red and blue light alternated at 2.5 kHz, while we synchronized the triggering of the respective blue and red laser diodes to their corresponding light patterns (Figure 4—figure supplement 1A,B). A 2.5 kHz switching rate was sufficient to avoid any temporal cross talk between the two DMD patterns and the on-rate and off-rate of the laser diodes.

For all experiments using the DMD (Figure 1 and 4), the photo-stimulation patterns were calibrated and aligned to the slice chamber. Prior to photo-stimulation, infrared and epifluorescence images were captured using an IR-1000 CCD camera (DAGE-MTI, Michigan City, Indiana) and imported into MATLAB. These images were used define the borders for photo-stimulation. The DMD was used to pattern light into a rectangular region that was 600 μm long in the dorso-ventral axis, and of variable length in the horizontal axis.

In all experiments using ramps of light to induce oscillations, blue light intensity was ramped from zero to a final intensity that was determined to maximize the power of inhibitory gamma oscillations in the recorded pyramidal cell, which was titrated prior to making recordings. The determined final blue light intensity varied slightly between slices and animals (1.0–1.3 mW mm2). Red light intensity for activation of eNpHR3.0 was 9.8 mW/mm2 in experiments using wide-field illumination (Figure 3), and at 3.6 mW/mm2 in experiments using DMD-based patterned illumination (Figure 4). For experiments in Figure 3—figure supplement 1, ChR2 was activated with a 1 mm multimode fiber-coupled blue LED mounted under the microscope objective, adjusted to illuminate all of V1. eNpHR3.0 was activated with a Hg+ lamp long pass filtered (>600 nm), and gated by a electromagentic shutter (Uniblitz) through a 40x objective (Olympus).

Data analysis

Request a detailed protocol

Analysis was performed in Matlab (Mathworks). Power spectra were computed using multi-tapered fourier estimation in Matlab with the Chronux package (http://chronux.org/)(Mitra and Bokil, 2008) using three tapers. Coherence was calculated as the magnitude-squared coherence using Welch’s overlapped averaged periodogram method. To obtain instantaneous phase information to correlate to the spike times of interneurons, a Hilbert transform was computed on the time series of inhibitory currents. Spike-oscillation coherence was measured using the pairwise phase consistency metric (PPC) described in Vinck et al. (2012). Synaptic charge during gamma activity was computed using trapezoidal integration of the voltage clamped currents at the corresponding reversal potentials for excitation (0 mV) and inhibition (−70 mV) after correction for the liquid junction potential (~7–10 mV) and baseline subtraction (mean of activity from t=-100 ms to t = 0 ms, relative to stimulus onset). For analysis in Figure 4—figure supplement 2C, current was integrated across the entire photo-stimulation period and normalized to the value at 0 μm. Membrane potential was computed by averaged the Vm during this same period. Maximum currents and depolarizations were computed by integrating currents or averaging Vm in the final 100 ms of the photo-stimulus.

Histology

Request a detailed protocol

To prepare histological sections, animals were anesthetized with a combination of ketamine (100 mg/kg) and xylazine (10 mg/kg) and perfused with cold PBS followed by cold 4% PFA. Brains were dissected and post-fixed for 2 hr at 4°C, rinsed 3 × 15 min in PBS, and cryopreserved for 24 hr in 30% sucrose in PBS at 4°C. Brains were then sectioned using a frozen microtome at a thickness of 40 μm. Confocal images of visual cortex were obtained and cells were manually counted using FIJI (ImageJ). For Figure 1—figure supplement 1: A top, B, and C, 400 μm acute slices used in physiology experiment were imaged.

References

  1. 1
  2. 2
  3. 3
  4. 4
  5. 5
  6. 6
  7. 7
  8. 8
  9. 9
  10. 10
  11. 11
  12. 12
  13. 13
  14. 14
  15. 15
  16. 16
  17. 17
  18. 18
  19. 19
  20. 20
  21. 21
    Columnar specificity of intrinsic horizontal and corticocortical connections in cat visual cortex
    1. CD Gilbert
    2. TN Wiesel
    (1989)
    The Journal of Neuroscience : The Official Journal of the Society for Neuroscience 9:2432–2442.
  22. 22
  23. 23
  24. 24
  25. 25
  26. 26
  27. 27
  28. 28
  29. 29
  30. 30
  31. 31
  32. 32
  33. 33
    Timing of action potentials and input currents in perisomatic and trilaminar interneurons during kainate induced gamma oscillations in mice CA1 hippocampus
    1. E Kipiani
    (2009)
    Georgian Medical News pp. 75–79.
  34. 34
  35. 35
  36. 36
  37. 37
  38. 38
  39. 39
  40. 40
  41. 41
  42. 42
    Observed Brain Dynamics
    1. P Mitra
    2. H Bokil
    (2008)
    Oxford; New York: Oxford University Press.
  43. 43
  44. 44
  45. 45
  46. 46
  47. 47
  48. 48
  49. 49
  50. 50
  51. 51
  52. 52
  53. 53
  54. 54
  55. 55
  56. 56
  57. 57
  58. 58
  59. 59
    Beta and gamma frequency synchronization by dendritic gabaergic synapses and gap junctions in a network of cortical interneurons
    1. J Szabadics
    2. A Lorincz
    3. G Tamás
    (2001)
    The Journal of Neuroscience : The Official Journal of the Society for Neuroscience 21:5824–5831.
  60. 60
  61. 61
  62. 62
  63. 63
  64. 64
    Gap junctions between interneuron dendrites can enhance synchrony of gamma oscillations in distributed networks
    1. RD Traub
    2. N Kopell
    3. A Bibbig
    4. EH Buhl
    5. FE LeBeau
    6. MA Whittington
    (2001)
    Journal of Neuroscience 21:9478–9486.
  65. 65
  66. 66
  67. 67
  68. 68
  69. 69
  70. 70
  71. 71
  72. 72

Decision letter

  1. Marlene Bartos
    Reviewing Editor; University of Freiburg, Germany

In the interests of transparency, eLife includes the editorial decision letter and accompanying author responses. A lightly edited version of the letter sent to the authors after peer review is shown, indicating the most substantive concerns; minor comments are not usually included.

Thank you for submitting your article "A neural circuit for long-range γ-band coherence in mouse visual cortex" for consideration by eLife. Your article has been reviewed by 3 peer reviewers, and the evaluation has been overseen by a Reviewing Editor and Timothy Behrens as the Senior Editor. The reviewers have opted to remain anonymous.

The reviewers have discussed the reviews with one another and the Reviewing Editor has drafted this decision to help you prepare a revised submission.

The study by Hakim and Adesnik examines the role of SOM-expressing interneurons in the synchronization of principal cell (PC) assemblies in the Layers 2/3 of the visual cortex slice preparation over large horizontal distances. All three reviewers judged the manuscript as important and interesting contribution to the field of neuroscience and as well performed. However, some concerns and questions were formulated from which the most important ones will be formulated in the following.

Previously, parvalbumin (PV)-expressing interneurons have been proposed to synchronize activity over large distances. The reviewers therefore ask for a discussion on the apparent dichotomy between the findings in this study on the synchronizing effect of SOM interneurons compared to PV cells. One of the reviewers asked whether due to the depressing nature of PV output synapses during repetitive activation and the facilitating characteristics of SOM interneuron outputs, γ oscillations generated at the onset of a strong sensory input may be PV mediated, while the sustained components later in the response could be SOM cell mediated. The reviewer therefore proposed to analyze γ expression using a wavelet method in sliding, 100 millisecond bins across the ramp stimulation period, and see if cell recruitment varies as a function of time period. Moreover, the authors need to be very clear in this manuscript over what period their analysis (e.g., of spiking rate, or phase locking) was conducted. The authors should weigh in on the question of why their results seem so different from prior studies in other brain areas. The authors conclude from their work that the oscillations are based on a PING mechanism. However, other mechanisms could also contribute such as gap junctional coupling between SOM cells and disynaptic inhibitory circuits involving VIP interneurons. The authors should mention these potentially involved mechanisms.

The study raised several questions related to the cellular mechanisms which may cause long-range synchronization which should be addressed in the Discussion of the manuscript: How do the SOM interneurons become synchronized to the far-away oscillation, and what are the limits on this? Are the local SOM interneurons receiving long-range excitatory input? Alternatively, do the SOM interneurons somehow transmit synchrony (e.g., through gap junctions) across the distance between the two sites? Or are sporadic excitatory neurons recruited at various points between the two sites? Do the SOM interneurons recruited at the site of the oscillation send long-range projections that inhibit the recorded pyramidal cell?

One of the reviewers was concerned regarding the precise location of the recorded PC in relation to the size of the applied light spot and intensity of light. Particularly for experiments in which a light spot was applied to one region and recorded a PC outside of that light spot, the distance between the edge of the light spot and the recorded PC was only ~200 µm. This suggests that inhibitory cells at the edge of the light spot might directly inhibit / entrain the recorded PC which does not qualify this as "long-distance" synchronization. To address this issue, the authors should record from the PC as a function of distance, preferably much farther from the edge of the light spot. A further important issue is that for experiments in which the authors looked at synchronization between two light spots, it would good to record from cells in between the two spots to examine how the two spots are communicating with each other. Since the main aim of the work was to examine the neuronal network underlying long-range synchronization of network oscillations, the experiments should be repeated as a function of distance.

Finally, the authors should quantify / characterize the frequency of the induced oscillations and mention this in the Abstract.

Reviewer #1:

This study examines the role of SOM interneurons in the synchronization of PC assemblies in Layer 2/3 of the visual cortex slice preparation over large horizontal distances. The authors provide evidence that long-range synchronization of γ activity patterns is driven by PCs projecting laterally and recruit local SOM cells (PING model). The study is overall well done. My main criticism relates to the question whether PV interneurons could support this long-range synchrony. In their previous work (Veit et al., 2017; Nat Neurosci) the authors tested the role of SOM cells in long-range synchrony in vivo and showed that optogenetic silencing of SOM cells reduces cross-correlation in Γ activity between both the local and more distant cortical sites. The effect of PV cells on long-range γ synchronization, however, was not tested. Moreover, is the role of SOM cells in long-range synchronization indeed layer-dependent? SOM cells are also located in other cortical layers but PCs do not project over large distances to other cortical regions (at least not as in layer II/III), similar results may not emerge in layer V. Please provide evidences for this assumption. Overall the manuscript is nice but it has the 'taste' of a secondary study of the Veit et al., 2017 paper.

Reviewer #2:

These experiments are well-conducted and will make an important contribution to the literature. They were cleanly designed to make the conceptual advancement argued for. There are several points that need to be addressed before publication:

I would recommend that the authors discuss the apparent dichotomy between their finding that SOM plays a critical role in γ generation, and previous literature attributing this role mainly to PV cells.

Could the difference be related to the integration window over which γ is measured? SOM neurons are known to show facilitating responses after a period of prior input, usually lasting tens to hundreds of milliseconds (but, certainly lasting at least a few γ cycles), whereas PYR → PV neurons can depress on the same time scale. If the authors only analyzed γ in the early phase of the response, is the disparity as great? I'm obviously guessing that early γ (soon after onset, say for the first 4 cycles) is PV mediated, while late γ (analyzed substantially after sensory/ChR2 onset, say 1+ seconds) is 'taken over' by the recruitment of SOM. If so, this temporal hand-off/evolution has interesting implications, and is a major step towards reconciling the prior relatively overwhelming literature showing γ mediated by FS-PV. In this vein, ramp-up stimulation to PCs is ideal for generating sparse activity across a pool of PCs initially that will drive PV suppression, and then as input gets stronger, optimal SOM recruitment. This comment is obviously (I hope!) not a critique of ramps, they likely replicate some kind of natural dynamic in the brain, but it does also help explain some of the differences in the literature and these current findings. Specifically, the γ generated at the onset of a strong sensory input – either bottom up salient or due to attention-mediated inhibition – may be PV mediated, while the sustained components later in the response (reflecting working memory, enhanced discriminative capability, etc.) could be SOMian.

• Recommendation I: Analyze γ expression using a wavelet method in sliding, 100 millisecond bins across the ramp stimulation period, and see if cell recruitment varies as a function of time period. I would also recommend inclusion here of such analysis applied to the recent Veit data generated with real sensory stimuli.

• Recommendation II: The authors need to be very clear in this manuscript over what period their analysis (e.g., of spiking rate, or phase locking) is conducted.

As a final, somewhat conceptual point: While the 'firewall' the authors place around their data as being about visual neocortex is fair, it implies that the differences from many prior papers in hippocampus and SI results from an area-of-the-brain issue, not an evolution across seconds issue (as raised above) or any of many other possibilities. As such, they should weigh in on the specific question of why their results seem so different from prior studies. Again, my guess is time frame, making interesting implications for function, but whatever their rumination, they should state it clearly.

Reviewer #3:

The subject is interesting, the experiments well done, and the slice preparation appropriate for answering detailed mechanistic questions. The main potential novelty here is the ability to elucidate the basic mechanisms of the oscillation beyond what this lab has already done in vivo, and to work out the mechanisms of long-range synchronization. With regard to working out mechanistic details of the oscillation, the main contribution is to observe the inhibitory and excitatory synaptic currents in pyramidal, SOM, and PV interneurons. The authors conclude that the oscillation reflects a PING mechanism; this seems likely although other mechanisms that could be at play here – specifically gap junctional coupling between SOM cells and disynaptic inhibitory circuits involving VIP interneurons. I'm not sure exactly what the authors could easily do to sort out these issues, but they are slightly tangential to the main point, so I think the authors could simply mention them.

Regarding the role of SOM interneurons in long-range synchrony, the authors conclude that long-range synchronization is driven by SOM interneurons. It is certainly true that if you induce an oscillation in one region of the slice, while recording from (and injecting depolarizing current into) a pyramidal cell some distance away, then when you inhibit SOM interneurons in the vicinity of the recorded pyramidal cell, its synchrony to the induced oscillation breaks down. Of course this sort of has to be the case – the L2/3 cells strongly activate SOM cells, which inhibit PV cells – most pyramidal cells seem to receive mainly inhibitory input, so if you disrupt the major source of inhibitory input to this cell, then it is effectively decoupled from the oscillation. The real question in my mind, is how do the SOM interneurons become synchronized to the far-away oscillation, and what are the limits on this? I can imagine many possibilities: Are the local SOM interneurons receiving long-range excitatory input? Alternatively, do the SOM interneurons somehow transmit synchrony (e.g., through gap junctions) across the distance between the two sites? Or are sporadic excitatory neurons recruited at various points between the two sites? Do the SOM interneurons recruited at the site of the oscillation send long-range projections that inhibit the recorded pyramidal cell? The authors dismiss the possibility that optogenetic inhibition affects terminals here, but I think we don't really know whether this is happening – the efficacy of such optogenetic inhibition on terminals likely depends on the details of cell type and preparation.

Furthermore, for these experiments, the Materials and methods specify that the width of the blue light region was 600 um, the width of the red light region was 350 um, there was a 150 μm gap between these two regions, and they patched cells in the middle of each region. This is quite different than the impression conveyed by the schematic in Figure 4 (and Figure 4—figure supplement 1) and I would recommend adjusting these figures to be more reflective of the actual experimental condition. Based on this, the distance between the patched cell and the edge of the blue-light induced oscillation is at most ~325 μm – likely less given that Figure 1 seems to suggest that oscillation-induced pyramidal cell spiking extends ~100 μm past the border of the blue light region. If the distance between the pyramidal cells directly excited by the local oscillation and the recorded "distant" pyramidal cell is only ~225 um, then I'm not sure this really qualifies as "long-distance" synchronization and I'm not sure the mechanisms underlying this experiment are necessarily the same ones that mediate the synchronization of oscillations induced by light delivery to two different, more widely regions of the slice. Notably, the width of the blue light region for this experiment was 600 μm whereas for the other experiment (two light patches to induce oscillations at two locations) it was 300 um. This raises the question of whether synchronization in this experiment vs. during stimulation of two different more widely separated regions are really the same phenomenon. In particular, when the patch of blue light is larger, the distance between the edge of the induced oscillation and the distant recorded cell will be much smaller than the distance between the center of the stimulated region and the "distant" recorded cell.

The title and Abstract suggest that the main motivation for this study is identifying the neural circuit that mediates long-range synchronization of oscillations. This is a very interesting topic that is certainly worthy of publication in eLife and the methods used are appropriate. Based on the issues raised above, I'm not sure the authors actually reveal this mechanism. That being said, I am enthusiastic about the direction of this work and think this concern could be addressed by repeating these experiments as a function of distance, and possibly making recordings from pyramidal cells and/or SOM interneurons in the intervening region.

Finally, the authors should quantify/characterize the frequency of the induced oscillations and be specific about this in the Abstract since "γ" oscillations mean different things to different people.

[Editors' note: further revisions were requested prior to acceptance, as described below.]

Thank you for resubmitting your work entitled "A neural circuit for γ-band coherence across the retinotopic map in mouse visual cortex" for further consideration at eLife. Your revised article has been favorably evaluated by Timothy Behrens (Senior Editor), a Reviewing Editor, and 1 reviewer.

The manuscript has been improved but there are some remaining issues that need to be addressed before acceptance, as outlined below:

The reviewers fully support the study and find that the phenomena outlined here are very interesting, important, and novel. However, the mechanistic understanding remains incomplete. Therefore additional experiments are required. The reviewer asks for one set of important experiments: record from SOM cells and pyramidal cells at various distances from a single patch of blue light / ChR2 excitation, and evaluate whether these cells receive long-distance EPSCs, long-distance IPSCs, and/or currents that appear to be gap-junction mediated. If SOM interneurons 400-500 µm away from the blue light patch receive EPSPs but do not show evidence of gap-junction mediated currents (e.g. spikelets) and there are not IPSPs in pyramidal neurons at a similar distance, then this would provide definitive evidence in favor of the author's proposed mechanism.

Reviewer #3:

I completely agree with the authors that:

1) The role of SOM interneurons in γ-frequency synchronization has long been underappreciated by the field;

2) SOM cells are essential for synchronization in the second experiment (shown in Figure 4).

The two key questions, which in my mind remain unanswered, are:

1) Does the same mechanism support long-range synchronization in the first experiment (Figure 1) and shorter-range synchronization in the second experiment (Figure 4)?

2) In the second experiment, what is the source of SOM-mediated synchronization – is it gap-junction coupling, medium-range SOM projections, or medium range excitatory input to local SOM interneurons?

Specifically, with regard to #2: the authors argue that gap-junction coupling is not sufficient to extend the range of synaptically evoked inhibition. That may be true, but it doesn't mean that SOM interneurons don't receive subthreshold rhythmic gap junction mediated currents that elicit synchronization. The authors argue against the possibility of medium-range SOM projections, by arguing that eNpHR would not be effective for silencing terminals, based on published data from Mahn et al. I would point out that the situation here is quite different from Mahn et al. – the distance between the edge of the red and blue light patches is only 150 um. Given that the red light seems to inhibit cells ~100 μm away (Figure 4S), and that the oscillation extends some distance beyond the edge of the blue light, it seems likely that the red light will directly suppress the activity of some SOM cells involved in the oscillation (not just inhibit their terminals).

So while the phenomena outlined here are very interesting, important, and novel, the mechanistic understanding remains incomplete. I am sympathetic to the experimental constraints noted by the authors. It may not be possible to repeat the second experiment using a smaller blue light patch/larger gap. However, given that this is a relatively short paper that builds on the author's previous work, my opinion is that there should be some test of how synchrony is transmitted by SOM cells. This could be quite minimal – I think simply recording from SOM cells and pyramidal cells at various distances from a single patch of blue light / ChR2 excitation, and evaluating whether these cells receive long-distance EPSCs, long-distance IPSCs, and/or currents that appear to be gap-junction mediated, would be very informative. I.e., if SOM interneurons 400-500 μm away from the blue light patch receive EPSPs but do not show evidence of gap-junction mediated currents (e.g., spikelets), and there are not IPSPs in pyramidal neurons at a similar distance, then this would provide definitive evidence in favor of the author's proposed mechanism.

The long-range transmission of synchrony (potentially) by SOM neurons is the key novel finding that differentiates this from the author's previous work, which is why I think this kind of additional evidence is important. If the authors were to find, for example, contrary to the preceding, that SOM cells at the border of the illuminated region (e.g., 50 μm beyond the edge of the blue light) provide inhibition to pyramidal cells ~200-300 μm away (very plausible given Figure 4D of Fino and Yuste), and SOM interneurons >300 μm away from the illuminated region do not receive significant EPSPs, then this would call into question the authors' interpretation.

https://doi.org/10.7554/eLife.28569.014

Author response

Reviewer #1:

This study examines the role of SOM interneurons in the synchronization of PC assemblies in Layer 2/3 of the visual cortex slice preparation over large horizontal distances. The authors provide evidence that long-range synchronization of γ activity patterns is driven by PCs projecting laterally and recruit local SOM cells (PING model). The study is overall well done. My main criticism relates to the question whether PV interneurons could support this long-range synchrony. In their previous work (Veit et al., 2017; Nat Neurosci) the authors tested the role of SOM cells in long-range synchrony in vivo and showed that optogenetic silencing of SOM cells reduces cross-correlation in Γ activity between both the local and more distant cortical sites. The effect of PV cells on long-range γ synchronization, however, was not tested. Moreover, is the role of SOM cells in long-range synchronization indeed layer-dependent? SOM cells are also located in other cortical layers but PCs do not project over large distances to other cortical regions (at least not as in layer II/III), similar results may not emerge in layer V. Please provide evidences for this assumption. Overall the manuscript is nice but it has the 'taste' of a secondary study of the Veit et al., 2017 paper.

While we agree that a role for PV neurons cannot be entirely excluded in long range synchronization in our slices, and have added this to the Discussion, there are several pieces of data that make this possibility significantly less likely. First, when we optogenetically inactivated SOM neurons and looked at distal synchronization, the phase locking of the target pyramidal cells was very strongly reduced, and in most neurons essentially abolished, implying that SOM neuron activity alone is sufficient for the γ rhythmicity (Figure 4). Second, when we generate large, coherent γ oscillations across the entire slice with broad illumination, PV neurons are only very weakly recruited, while SOM neurons are very strongly recruited (Figure 2). If PV cells were critical for long range synchrony, one would expect PV neurons to be more strongly recruited. These points have been added to the Discussion as well. Third, under these same conditions where γ oscillations are coherent across nearly all of V1 in the brain slice (Figure 1), suppressing PV neurons has no significant effect on γ. Under all these conditions where long-range synchronization is very prominent, either PV neurons are inactive, not necessary, or SOM neurons account for nearly all of the observed coherence or γ power.

With respect to layer-specific recruitment of SOM cells in long-range γ entrainment: This is an important and fascinating question that we address more thoroughly in the Discussion, but requires further study in the future. In this work, we chose to constrain our focus to L2/3-generated γ oscillations for simplicity and clarity, and while our results point to critical roles of L2/3 SOM cells in the oscillations that we study, we do not exclude a role for SOM cells in other layers, or that oscillations generated in other layers may depend on other inhibitory neuron subtypes. One line of evidence strongly suggesting that L2/3 SOM cells, and not SOM cells in other layers (particularly L5 Martinotti cells), are mediating the long-range coherence is the experiment in Figure 1G-J, which shows the necessity of lateral connections in only L2/3 on long-range coherence. However, Kapfer et al. (2007) show that recurrent connections between L2/3 PCs and L5 Martinotti cells, suggesting that local rhythm generation may involve deeper layer SOM cells as well.

Reviewer #2:

These experiments are well-conducted and will make an important contribution to the literature. They were cleanly designed to make the conceptual advancement argued for. There are several points that need to be addressed before publication:

I would recommend that the authors discuss the apparent dichotomy between their finding that SOM plays a critical role in γ generation, and previous literature attributing this role mainly to PV cells.

We have substantially expanded on this point in the Discussion. It takes very minimal circuitry to generate γ rhythms (Sohal et al. 2009; Veit et al. 2017 (Figure 4)), and neural circuits in the cortex, hippocampus, and other areas are composed of a sufficiently large number of distinct cell types to easily support different γ generating mechanisms in different layers, subnetworks, or conditions. We now emphasize that in many circuits, both in vivo and in vitro, PV neurons are clearly shown to fire and synchronize to γ rhythms. However, we also point out that entrainment to γ and a role in its generation are potentially distinct. For instance, as Veit et al. recently demonstrated, PV neurons in the mouse visual cortex lock well to visually induced γ, but this rhythm is abolished by suppressing SOM and not PV neurons. This result implies that in studies where PV neurons locked to γ, the underlying oscillation might still be dependent on SOM and not PV neurons. Distinguishing between these possibilities requires causal manipulations, such as optogenetic suppression of SOM cells. However, in a few cases, causal manipulations of PV neurons have demonstrated that they are either necessary or sufficient for γ entrainment, such as in the barrel cortex (Cardin et al.) or in the prefrontal cortex (Sohal et al.). Our conclusion is that long-range synchronization in L2/3 of the visual cortex (at least in mice) is dependent more intimately on SOM neurons, but that γ rhythms in other brain areas, or even other layers may instead depend on PV neurons or alternative circuits. Since long-range synchronization in superficial V1 is one of the best studied models for γ rhythms, we consider our study impactful, whether or not the same mechanism may hold for the various γ rhythms observed in the hippocampus or prefrontal cortex.

Could the difference be related to the integration window over which γ is measured? SOM neurons are known to show facilitating responses after a period of prior input, usually lasting tens to hundreds of milliseconds (but, certainly lasting at least a few γ cycles), whereas PYR → PV neurons can depress on the same time scale. If the authors only analyzed γ in the early phase of the response, is the disparity as great? I'm obviously guessing that early γ (soon after onset, say for the first 4 cycles) is PV mediated, while late γ (analyzed substantially after sensory/ChR2 onset, say 1+ seconds) is 'taken over' by the recruitment of SOM. If so, this temporal hand-off/evolution has interesting implications, and is a major step towards reconciling the prior relatively overwhelming literature showing γ mediated by FS-PV. In this vein, ramp-up stimulation to PCs is ideal for generating sparse activity across a pool of PCs initially that will drive PV suppression, and then as input gets stronger, optimal SOM recruitment. This comment is obviously (I hope!) not a critique of ramps, they likely replicate some kind of natural dynamic in the brain, but it does also help explain some of the differences in the literature and these current findings. Specifically, the γ generated at the onset of a strong sensory input – either bottom up salient or due to attention-mediated inhibition – may be PV mediated, while the sustained components later in the response (reflecting working memory, enhanced discriminative capability, etc.) could be SOMian.

• Recommendation I: Analyze γ expression using a wavelet method in sliding, 100 millisecond bins across the ramp stimulation period, and see if cell recruitment varies as a function of time period. I would also recommend inclusion here of such analysis applied to the recent Veit data generated with real sensory stimuli.

The reviewer proposes an intriguing hypothesis that we have now addressed through additional experiments and analyses. A temporal ‘hand-off’ in γ mediation from PV to SOM cells would indeed be revealing. To address this, first, we performed the analysis suggested by the reviewer (Figure 3—figure supplement 2A-E) by aligning SOM and PV spike times to the onset of IPSCs. Alignment to the onset of IPSCs provided a more consistent and interpretable metric than alignment to the onset time of γ power using wavelet analysis, though a similar aggregate result emerges with this method. We find that PV spiking is sparse during all epochs of γ and reaches peak activity after SOM cells do (300-400ms for PV, 100-200ms for SOM). Second, we repeated the experiment in Figure 3, but initiated SOM suppression prior to the onset of inducing γ. We found that γ power remains abolished (Figure 3—figure supplement 2F-H). These experiments demonstrate that SOM neuron activity is required for both the early and ongoing phase of layer 2/3 γ oscillations and that PV neurons are neither necessary nor specifically recruited for initiation or maintenance under our conditions. This said, we agree (and now add to the Discussion) that different temporal patterns of photo-stimulation (other than ramps) could potentially lead to conditions where PV neurons should fire at the onset of the response (as they do with a brief pulse of light – see Figure 3—figure supplement 1).

Consistent with well-known results from the hippocampus and Layer 4 of the barrel cortex, L2/3 PV neurons clearly respond well to brief, strong increases in afferent input, while SOM cells responds better to more sustained input. Thus while SOM cells appear sufficient in our conditions to initiate and maintain γ oscillations without much input from PV neurons, under ‘natural conditions’ – such as time varying input during natural vision, we would expect that PV neurons are periodically recruited, and their inhibitory output may be involved in phase modulating the γ rhythm on a cycle by cycle basis. This remains to be explored in the future. In Veit et al., 2017, we also showed that photo-suppression of SOM cells from before visual stimulus onset likewise impaired γ induction, demonstrating that SOM cells are involved even at the very earliest stages of γ entrainment. Similar photo-suppression of PV cells did not have the same effect, although we must cautiously interpret this experiment since it was not possible to suppress PV cells very strongly in awake mice without inducing epileptic events. Hence, in that paper we conclude that both SOM and PV cells are required for γ oscillations, but that PV cells’ role in this γ is primarily for network stabilization, while the SOM cells are more intimately involved in γ entrainment.

• Recommendation II: The authors need to be very clear in this manuscript over what period their analysis (e.g., of spiking rate, or phase locking) is conducted.

Additional text has been added to clearly define the time period over which our analyses are conducted. In experiments where interneuron subtypes were inhibited during γ (Figure 3, Figure 3—figure supplement 1, Figure 3—figure supplement 2F-H, Figure 4), analysis was generally conducted during epochs where the red light was on or would have been on in complementary control trials: 250ms-750ms for Figure 3, Figure 3—figure supplement 1; 0-1000ms for Figure 3—figure supplement 2F-H; and in Figure 4 we analyzed from 200-1000ms so as to ignore spikes that occurred prior to the emergence of steady oscillations and after the adapting period of regular-spiking L2/3 Pyramidal cells. In Figure 4—figure supplement 1, we analyzed during the red light epoch in control and red light trials (250-750 ms). For all other experiments (Figure 1, Figure 1—figure supplement 2, Figure 2) the entire stimulation period (0-1000ms) was used for analysis.

As a final, somewhat conceptual point: While the 'firewall' the authors place around their data as being about visual neocortex is fair, it implies that the differences from many prior papers in hippocampus and SI results from an area-of-the-brain issue, not an evolution across seconds issue (as raised above) or any of many other possibilities. As such, they should weigh in on the specific question of why their results seem so different from prior studies. Again, my guess is time frame, making interesting implications for function, but whatever their rumination, they should state it clearly.

We agree this is a critical point for discussion. As we stated above, we have substantially amended the Discussion and address the issue of temporal dynamics. It is true that many studies have found that PV or fast spiking (presumed PV neurons) typically lock to ongoing γ, and in a small number of studies, have been causally related to γ. However, we emphasize that locking to γ proves neither causality nor sufficiency, as PV neurons receive powerful inhibition from SOM neurons and can be entrained by SOM cells themselves, as we recently showed during visually induced γ rhythms in mouse V1 (Veit et al., 2017). Thus, we suspect that if this question were reexamined more broadly, a subset of the instances of γ in other brain circuits might also be found to depend on SOM neurons.

Reviewer #3:

The subject is interesting, the experiments well done, and the slice preparation appropriate for answering detailed mechanistic questions. The main potential novelty here is the ability to elucidate the basic mechanisms of the oscillation beyond what this lab has already done in vivo, and to work out the mechanisms of long-range synchronization. With regard to working out mechanistic details of the oscillation, the main contribution is to observe the inhibitory and excitatory synaptic currents in pyramidal, SOM, and PV interneurons. The authors conclude that the oscillation reflects a PING mechanism; this seems likely although other mechanisms that could be at play here – specifically gap junctional coupling between SOM cells and disynaptic inhibitory circuits involving VIP interneurons. I'm not sure exactly what the authors could easily do to sort out these issues, but they are slightly tangential to the main point, so I think the authors could simply mention them.

We have clarified that our statement that a PING mechanism likely underlies γ generation is a hypothesis that is suggested but not proven by our data, and as such, additional discussion on other plausible mechanisms, such as the role of electrical connections has been included.

Regarding the role of SOM interneurons in long-range synchrony, the authors conclude that long-range synchronization is driven by SOM interneurons. It is certainly true that if you induce an oscillation in one region of the slice, while recording from (and injecting depolarizing current into) a pyramidal cell some distance away, then when you inhibit SOM interneurons in the vicinity of the recorded pyramidal cell, its synchrony to the induced oscillation breaks down. Of course this sort of has to be the case – the L2/3 cells strongly activate SOM cells, which inhibit PV cells – most pyramidal cells seem to receive mainly inhibitory input, so if you disrupt the major source of inhibitory input to this cell, then it is effectively decoupled from the oscillation. The real question in my mind, is how do the SOM interneurons become synchronized to the far-away oscillation, and what are the limits on this? I can imagine many possibilities: Are the local SOM interneurons receiving long-range excitatory input? Alternatively, do the SOM interneurons somehow transmit synchrony (e.g., through gap junctions) across the distance between the two sites? Or are sporadic excitatory neurons recruited at various points between the two sites? Do the SOM interneurons recruited at the site of the oscillation send long-range projections that inhibit the recorded pyramidal cell? The authors dismiss the possibility that optogenetic inhibition affects terminals here, but I think we don't really know whether this is happening – the efficacy of such optogenetic inhibition on terminals likely depends on the details of cell type and preparation.

The reviewer brings up several important points that we address here and by modifying the Discussion. First, we agree with the reviewer’s intuition that the distal entrainment of spike-timing to fast oscillations by SOM neurons could potentially be predicted by what we know of cortical circuitry; however this prediction has not been experimentally demonstrated and has not been explicated as a prediction by many leaders in the field (Traub et al., 1996c, Bartos et al., 2007, Buzsaki and Wang, 2012). Therefore, we feel that demonstrating this result experimentally has important value.

Second, SOM neurons do receive significant long range excitatory input, as shown by experiments in Adesnik et al., 2012. We believe this long range excitatory input to SOM cells from L2/3 pyramidal cells is at the heart of the mechanism for γ synchronization under study here.

With respect to the possibility that SOM cells transmit synchrony via gap junctions: This is an intriguing idea that has some support both at the single cell level (Hu and Agmon, 2015) and network level (Connors, 2017). However, if we compare the spatial fall-off of pair-wise connectivity from SOMs onto PCs (Fino and Yuste, 2011) vs. the spatial fall-off of IPSC amplitudes recorded in PCs by stimulating populations of SOM cells at different distances (Kato et al., 2017), we see essentially identical spatial fall-off curves. This suggests that, at a network level, in layer 2/3 of cortical slices, SOM cells do not receive a significant boost to their spatial propagation due to gap-junctions, although this could be tested experimental in the future with appropriate methods to block or delete gap junctions in SOM cells.

With respect to the possibility that sporadic excitatory neurons are recruited at various points between two distantly-separated ensembles: In agreement with previous work demonstrating that layer 2/3 Pyramidal cells located outside the blue light stimulation receive strong lateral inhibition mediated by SOM cells and are net suppressed (Adesnik et al., 2010, 2012), we never found pyramidal cells outside the blue light illumination zone that were synaptically driven to spike (Figure 1—figure supplement 2). In the experiment in Figure 4, all recorded Pyramidal cells outside the blue light region required current injection (~300-600 pA, roughly 100-400 pA above typical rheobase (Guan et al., 2007; van der Velden et al., 2012; Lefort et al., 2008)) in order to induce spiking. In other words, our data indicate that the only L2/3 excitatory neurons that are spiking in the slice are those that are in the blue light regions, implying that those in between are not firing due to synaptic activation. This essentially rules out the possibility that the synchronization is propagated through multisynaptic loops across the slice.

As to whether SOM cells send long range projections, McGarry et al., 2010 and Fino and Yuste, 2011 found that a subset of SOM neurons do exhibit axons that ascend towards layer 1, make a 90 degree turn, and then project horizontally from anywhere between 100 and 400 hundred microns. Therefore, the reviewer is correct that if our red light stimulation of eNpHR3.0 did suppress synaptic release from SOM terminals, a possible interpretation is that these horizontally-projecting axons were critical. Definitively testing whether this was the case under our conditions has proven to be very tricky (experiments with co-expression of ChR2 and eNpHR3.0 in our lab have always ended up inconclusive). Nevertheless, the definitive study on this methodological problem, so far, is Mahn et al., Nature Neuroscience 2016. Figure 2B of that paper shows that illumination of eNpHR3.0 expressing terminals suppresses the response to the first spike in a two spike sequence, but potentiates the second. In other words, optogenetic suppression of terminals appears only to reduce the probability of synaptic release, but not suppress transmitter release under conditions of sustained presynaptic activity (as studied here, and as is typical in most in vivo scenarios). While this study did not examine SOM cells’ terminals specifically, we surmise that the same might be true, and that our red light illumination would only have very transiently suppressed GABA release from SOM cell axons. This suggests that axon terminals were not specifically inhibited in our experiment, and those horizontally projecting SOM cells maintained their output onto the recorded Pyramidal cell with and without red light. Please also note that under Mahn’s conditions the average suppression was only ~20% on the first synaptic (Figure 2H). The dramatic example they show in Figure 2B is clearly not representative. Thus we conclude that terminal suppression with eNpHR3.0 is actually quite weak and largely ineffective during sustained activity. Nevertheless, we cannot yet rule this contingency out with certainty, and we bring this up as an important issue in the Discussion.

Furthermore, for these experiments, the Materials and methods specify that the width of the blue light region was 600 um, the width of the red light region was 350 um, there was a 150 μm gap between these two regions, and they patched cells in the middle of each region. This is quite different than the impression conveyed by the schematic in Figure 4 (and Figure 4—figure supplement 1) and I would recommend adjusting these figures to be more reflective of the actual experimental condition. Based on this, the distance between the patched cell and the edge of the blue-light induced oscillation is at most ~325 μm – likely less given that Figure 1 seems to suggest that oscillation-induced pyramidal cell spiking extends ~100 μm past the border of the blue light region. If the distance between the pyramidal cells directly excited by the local oscillation and the recorded "distant" pyramidal cell is only ~225 um, then I'm not sure this really qualifies as "long-distance" synchronization and I'm not sure the mechanisms underlying this experiment are necessarily the same ones that mediate the synchronization of oscillations induced by light delivery to two different, more widely regions of the slice. Notably, the width of the blue light region for this experiment was 600 μm whereas for the other experiment (two light patches to induce oscillations at two locations) it was 300 um. This raises the question of whether synchronization in this experiment vs. during stimulation of two different more widely separated regions are really the same phenomenon. In particular, when the patch of blue light is larger, the distance between the edge of the induced oscillation and the distant recorded cell will be much smaller than the distance between the center of the stimulated region and the "distant" recorded cell.

The title and Abstract suggest that the main motivation for this study is identifying the neural circuit that mediates long-range synchronization of oscillations. This is a very interesting topic that is certainly worthy of publication in eLife and the methods used are appropriate. Based on the issues raised above, I'm not sure the authors actually reveal this mechanism. That being said, I am enthusiastic about the direction of this work and think this concern could be addressed by repeating these experiments as a function of distance, and possibly making recordings from pyramidal cells and/or SOM interneurons in the intervening region.

We have amended the schematics in Figure 4 as per the reviewer’s recommendations.

In the experiment in Figure 4, we aimed to confirm the conclusion that SOM cells mediate γ entrainment across the horizontal axis. We originally tried separating the two regions (blue patch and spiking Pyramidal cell/red patch) by distances as long as in Figure 1 (up to ~800um), as the reviewers suggests, however, we found that the net synaptic input to the target pyramidal cell was weak and γ entrainment was not robust compared to that seen in Figure 1 or the shorter edge-to-edge distance that we ultimately used in Figure 4. The best explanation for this is that the strong coherence seen at long distances in Figure 1 is because both sites get blue light photo-stimulation, so that interneurons in both patches receive both local input from nearby pyramidal cells (which is relatively stronger and brings them to action potential threshold), and a smaller amount of long-range input from the other blue light patch. Thus we concluded that while the long-range input, on its own, is insufficient to bring interneurons to AP threshold, it is nevertheless sufficient to phase lock the two independently oscillating ensembles into coherence when they are both driven with blue light. Conversely, in the experiment in Figure 4, since only one patch gets direct photo-stimulation with blue light, the interneurons in the red light patch are only receiving distal input, which, when restricted to long-range distances, is insufficient on its own to bring local [SOM] interneurons to action potential threshold. Hence, we used a wider blue light patch that includes both long-range (up to ~950μm) and more proximal regions of L2/3 (~200μm) (see updated Figure 4 and Figure 4 supplement). Taking these constraints into consideration, this configuration tests the necessity of SOM cell activity on the sum of more proximal and long-range horizontal input, rather than just long-range input alone. Given the long-range coherence and recruitment we observe in Figure 1, as well as literature on connectivity as a function of distance (Fino and Yuste, 2011; Kato et al. 2017), we surmise that both nearby and long-range inputs contribute to phase coherence in the recorded spiking Pyramidal cell in Figure 4.

Upon suppressing local SOM cells (within ~200 μm), we find coherence is abolished. Critically, this experiment demonstrates that whether it be local SOM cells or distally-projecting SOM cells, it is indeed SOM cells mediating both nearby and long-range γ entrainment. If other interneuron subtypes or direct excitation meaningfully contributed to nearby or long-range entrainment, spike-oscillation coherence upon red light illumination would remain to some extent, yet we see it abolished nearly completely. This result solidifies our core hypothesis that SOM cells mediate long-range γ synchronization, and is discussed more clearly in the revised manuscript.

Finally, the authors should quantify/characterize the frequency of the induced oscillations and be specific about this in the Abstract since "γ" oscillations mean different things to different people.

We have added this information to the Abstract.

[Editors' note: further revisions were requested prior to acceptance, as described below.]

The manuscript has been improved but there are some remaining issues that need to be addressed before acceptance, as outlined below:

The reviewers fully support the study and find that the phenomena outlined here are very interesting, important, and novel. However, the mechanistic understanding remains incomplete. Therefore additional experiments are required. The reviewer asks for one set of important experiments: record from SOM cells and pyramidal cells at various distances from a single patch of blue light / ChR2 excitation, and evaluate whether these cells receive long-distance EPSCs, long-distance IPSCs, and/or currents that appear to be gap-junction mediated. If SOM interneurons 400-500 µm away from the blue light patch receive EPSPs but do not show evidence of gap-junction mediated currents (e.g. spikelets) and there are not IPSPs in pyramidal neurons at a similar distance, then this would provide definitive evidence in favor of the author's proposed mechanism.

We are glad the reviewers fully support the study, and we have conducted the proposed experiments. Our data confirms the three predictions mentioned above: First, at 450 µm SOM cells get appreciable (though reduced) excitatory currents. Second, pyramidal cells get even less inhibition. Third, we observed no obvious spikelets or evidence of propagation of subthreshold potentials in SOM cells that would indicate propagations of signals between electrically coupled SOM cells across the retinotopic axis. We discuss these points in detail below.

Reviewer #3:

I completely agree with the authors that:

1) The role of SOM interneurons in γ-frequency synchronization has long been underappreciated by the field;

2) SOM cells are essential for synchronization in the second experiment (shown in Figure 4).

The two key questions, which in my mind remain unanswered, are:

1) Does the same mechanism support long-range synchronization in the first experiment (Figure 1) and shorter-range synchronization in the second experiment (Figure 4)?

2) In the second experiment, what is the source of SOM-mediated synchronization – is it gap-junction coupling, medium-range SOM projections, or medium range excitatory input to local SOM interneurons?

Fully discussed below, but the new circuit mapping data implies that synchronization in slice beyond 300 µm benefits from convergent excitatory input from distant excitatory ensembles on both sides to drive individual SOM cells. While the extent of horizontal excitatory projections is surely critical, the spatial spread of synchronization is likely expanded by the horizontal spread of SOM axons themselves, as suggested by the reviewer.

Specifically, with regard to #2: the authors argue that gap-junction coupling is not sufficient to extend the range of synaptically evoked inhibition. That may be true, but it doesn't mean that SOM interneurons don't receive subthreshold rhythmic gap junction mediated currents that elicit synchronization.

We agree with this point and have amended the text. We recorded SOM cells in current clamp while stimulating increasingly more distant ensembles of excitatory neurons, but never observed obvious spikelets. This doesn’t rule out a contribution of gap junctions, since spikelets might appear temporally filtered to an extent they are indistinguishable or buried in the large barrages of glutamatergic EPSPs (Hu and Agmon, 2015), and the text has been updated to reflect this. With respect to subthreshold rhythmic gap junction currents, this is also not something we can entirely rule out. However, if subthreshold electrical propagation contributed to SOM cell spike-timing, we might expect that we would observe a filtered version of what a SOM neuron receives for more proximal stimulation (that is, a low pass filtered version of EPSPs). Instead, membrane current traces in SOM neurons show clear periods of baseline/quiescence between what appear to be unitary-like synaptic conductances or EPSPs in response to distal excitation (Figure 4—figure supplement 2E, G); traces during more proximal stimulation show no such quiescent periods. This is not consistent with substantial propagation of subthreshold electrical inputs from proximal to distal SOM cells.

Given the significant off-target effects of all gap-junction blocking drugs (Juszczak and Swiergiel, 2009), future work in mouse lines with connexin knock outs could more thoroughly address this point. Although there is little work on this topic, genetic deletion of connexin36 (Cx36) largely abolishes electrical coupling between cortical interneurons (Deans et al., Neuron, 2001), and thus could be used as a reasonably definitive test for the requirement of electrical coupling in the light evoked γ. These authors have previously presented data (Connors et al., SFN Abstract, 2014, “Roles of electrical synapses in γ-band activity of sensory and association cortex”) that showed that light-ramp evoked γ oscillations in the sensory cortex are observed in Cx36 KO mice, although these γ oscillations were in the barrel cortex and these authors used an AAV with a CamKII promoter, which is not selective to L2/3. However, it should be possible to test the contribution of gap junctions to this γ coherence definitively by electroporating ChR2 into V1 of Cx36 KOs – an interesting experiment we think is appropriate for a future study.

The authors argue against the possibility of medium-range SOM projections, by arguing that eNpHR would not be effective for silencing terminals, based on published data from Mahn et al. I would point out that the situation here is quite different from Mahn et al. – the distance between the edge of the red and blue light patches is only 150 um. Given that the red light seems to inhibit cells ~100 μm away (Figure 4S), and that the oscillation extends some distance beyond the edge of the blue light, it seems likely that the red light will directly suppress the activity of some SOM cells involved in the oscillation (not just inhibit their terminals).

We agree with this point as well, and have amended the text. As mentioned above, our data indicate that the spatial spread of L2/3 excitatory axons is critical, but the spatial spread of SOM cell axons themselves are likely to extend the range of synchronization. In Figure 4, as the reviewer points out, some of the drop in synchronization is most likely a result of direct somato-dendritic suppression of SOM cells that lie between the blue and red light zones. The text has been amended to reflect this fact.

So while the phenomena outlined here are very interesting, important, and novel, the mechanistic understanding remains incomplete. I am sympathetic to the experimental constraints noted by the authors. It may not be possible to repeat the second experiment using a smaller blue light patch/larger gap. However, given that this is a relatively short paper that builds on the author's previous work, my opinion is that there should be some test of how synchrony is transmitted by SOM cells. This could be quite minimal – I think simply recording from SOM cells and pyramidal cells at various distances from a single patch of blue light / ChR2 excitation, and evaluating whether these cells receive long-distance EPSCs, long-distance IPSCs, and/or currents that appear to be gap-junction mediated, would be very informative. I.e., if SOM interneurons 400-500 μm away from the blue light patch receive EPSPs but do not show evidence of gap-junction mediated currents (e.g., spikelets), and there are not IPSPs in pyramidal neurons at a similar distance, then this would provide definitive evidence in favor of the author's proposed mechanism.

We have conducted the proposed experiments. The data is presented as Figure 4—figure supplement 2. These experiments show that SOM cells do receive appreciable current as far away as 450 µm. While the absolute amount of excitatory synaptic charge at this distance dropped to ~6% of the maximum, the max synaptic current still averaged 120 ± 90 pA, and several of the recorded SOM cells exhibited depolarizations of 10-20 mV, and occasionally, though rarely, could even be driven to spike. The large depolarizations (and some evoked spiking), despite the reduced currents, are likely due to the relatively high membrane resistance of SOM cells (also note that even 6% of the very strong input seen at 0 µm may still be sufficient input to facilitate synchronization with other active ensembles even if it does not spike the SOM neurons on its own). At these same distances we observed no obvious spikelets (see expanded voltage- and current-clamp traces in Figure 4—figure supplement 2), although these cannot be entirely ruled out. We also quantified inhibitory synaptic input in pyramidal cells as a function of distance. The overall shape of this curve compared to that of synaptic excitation to SOM neurons is similar (consistent with Adesnik and Scanziani, 2010, in the barrel cortex), although it may fall off slightly faster in space as the reviewer predicts. Perhaps more importantly, the absolute amplitude of the inhibitory currents were very small (max 20 ± 10 pA, and sometimes non-existent at 450 µm).

We conclude from these data that the loss of synchronization observed in Figure 4 (entrainment in SOM cell at ~300-400 µm from blue light patch) is likely due to somatodendritic suppression of SOM cells that are mostly located near, but not necessarily in, the red light zone. SOM cells in the red light zone itself are substantially depolarized by horizontal input, and a few of them are even driven to spike, but probably don’t contribute as much as those SOM cells that lie between the two light patches. The Discussion has been updated to reflect this point.

In Figure 1, when we observed γ coherence between ensembles separated by as much as 800 µm, we propose the scenario is somewhat different (and potentially more similar to what is going in the brain during sensory stimulation with multiple stimuli). Here, two spatially separated ensembles of excitatory neurons – each driven independently to oscillate with their own patch of blue light – converge on SOM cells across the horizontal axis, but particularly in between them. The combined input would drive these SOM cells to spike. Thus, SOM cells across the horizontal axis are recruited by summating input from the two sides, and the combined action of all these SOM cells help synchronize the distant ensembles. Despite the absence of observed spikelets or electrically coupled subthreshold potentials, we don’t rule out gap junctional coupling as a contributor to this process, but assessing the specific role of gap junctions will require future work in CXx36 animals.

The long-range transmission of synchrony (potentially) by SOM neurons is the key novel finding that differentiates this from the author's previous work, which is why I think this kind of additional evidence is important. If the authors were to find, for example, contrary to the preceding, that SOM cells at the border of the illuminated region (e.g., 50 μm beyond the edge of the blue light) provide inhibition to pyramidal cells ~200-300 μm away (very plausible given Figure 4D of Fino and Yuste), and SOM interneurons >300 μm away from the illuminated region do not receive significant EPSPs, then this would call into question the authors' interpretation.

As described above, based on the new data, we have appropriately revised our proposed mechanism. All these data are now in the manuscript, and a full discussion of its interpretation has been added to the Discussion. In brief, SOM interneurons >300 μm away from the illuminated region do get significant synaptic input that could facilitate γ entrainment, but we have updated the text to reflect the fact that the spatial spread of SOM neurons’ axons themselves most likely contributes to the propagation of synchrony as well. With these modifications, we feel that the current interpretation as it stands in the revised manuscript is fully supported by the data.

https://doi.org/10.7554/eLife.28569.015

Article and author information

Author details

  1. Richard Hakim

    1. Department of Molecular and Cell Biology, University of California, Berkeley, United States
    2. Helen Wills Neuroscience Institute, University of California, Berkeley, United States
    Present address
    Department of Neurobiology, Harvard Medical School, Massachusetts, United States
    Contribution
    Conceptualization, Resources, Supervision, Funding acquisition, Writing—original draft, Project administration, Writing—review and editing
    Competing interests
    No competing interests declared
    ORCID icon "This ORCID iD identifies the author of this article:" 0000-0002-6991-1801
  2. Kiarash Shamardani

    1. Department of Molecular and Cell Biology, University of California, Berkeley, United States
    2. Helen Wills Neuroscience Institute, University of California, Berkeley, United States
    Contribution
    Investigation, Methodology
    Competing interests
    No competing interests declared
  3. Hillel Adesnik

    1. Department of Molecular and Cell Biology, University of California, Berkeley, United States
    2. Helen Wills Neuroscience Institute, University of California, Berkeley, United States
    Contribution
    Resources, Methodology, Project administration
    For correspondence
    hadesnik@berkeley.edu
    Competing interests
    No competing interests declared
    ORCID icon "This ORCID iD identifies the author of this article:" 0000-0002-3796-8643

Funding

National Eye Institute (R01EY023756-01)

  • Hillel Adesnik

New York Stem Cell Foundation

  • Hillel Adesnik

The funders had no role in study design, data collection and interpretation, or the decision to submit the work for publication.

Acknowledgements

The authors thank J Veit for extensive discussions, D Taylor and A Naka for technical assistance, and J Veit, B Sabatini, J Assad, and R Born for a critical reading of the manuscript. This work was supported by NEI grant R01EY023756-01 and the New York Stem Cell Foundation. HA is a New York Stem Cell - Robertson Investigator.

Ethics

Animal experimentation: All experiments were performed in accordance with the guidelines and regulations of the ACUC of the University of California, Berkeley and the IACUC of the University of California, San Diego. Protocol # AUP-2014-10-6832

Reviewing Editor

  1. Marlene Bartos, University of Freiburg, Germany

Publication history

  1. Received: May 12, 2017
  2. Accepted: February 15, 2018
  3. Version of Record published: February 26, 2018 (version 1)

Copyright

© 2018, Hakim et al.

This article is distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use and redistribution provided that the original author and source are credited.

Metrics

  • 1,537
    Page views
  • 281
    Downloads
  • 3
    Citations

Article citation count generated by polling the highest count across the following sources: Scopus, PubMed Central, Crossref.

Download links

A two-part list of links to download the article, or parts of the article, in various formats.

Downloads (link to download the article as PDF)

Download citations (links to download the citations from this article in formats compatible with various reference manager tools)

Open citations (links to open the citations from this article in various online reference manager services)